The Real-Time Detection of Traffic Participants Using YOLO Algorithm

Abstract

Object detection is one of the key software components in the next generation of autonomous cars. Classical computer vision and machine learning approaches for object detection usually suffer from the slow response time. Modern algorithms and architectures based on artificial neural networks, such as YOLO (You Only Look Once) algorithm, solve this problem without precision losses. In this paper we provide the demonstration of the usage of the newest YOLOv3 algorithm for the detection of traffic participants. We have trained the network for 5 object classes (car, truck, pedestrian, traffic signs, and lights) and have demonstrated the effectiveness of the approach in the variety of the driving conditions (bright and overcast sky, snow, fog, and night).